One of the less commonly talked about things that would be potentially useful for improving things would be collecting the background noise of people’s environments. There’s tools to build denoising models out and about, and while we can rely on the stock models, building up enough samples to make a more representative mycroft model could be useful if/when denoising gets added to the audio pipeline. A couple of 3-5 second clips of your location without anyone speaking (or even with non-distinct speaking in the background) would be it. IF there’s significant variation from times of day, perhaps a sample from each relevant time. Since it’s not holding voice data, should be anonymous collection. Not a high priority by any means, maybe even add it as an option for a precise/wakeword collector tool.